How to give boost pool access to a gsl matrix for threading a task

Question

I have a gsl matrix that I'm filling over and over again and its pretty heavy to do it all on one thread. Is it possible to modify a gsl matrix in a boost pool thread?

#include <gsl/gsl_linalg.h>
#include <boost/asio.hpp>

int main() {
    int SIZE = 3; // small size
    gsl_matrix * _A = gsl_matrix_alloc(SIZE, SIZE);  // set the gsl matrix to 3x3
    gsl_matrix_set_all(_A, 0);  // make them all 0 to provide contrast if value changes
    auto _dostuff = [&_A]() {
        gsl_matrix_set (_A, 0, 0, 20); // set the (0, 0) position to 20
    };
    boost::asio::thread_pool pool(1);  // open up a thread pool containing 1 thread
    boost::asio::post(pool, [&_A, &_dostuff]{_dostuff() ;});  // post a task to the pool
    std::cout << gsl_matrix_get (_A, 0, 0) << std::endl;  // check to see if the (0, 0) position is 20 (its not)

    return 0;
}

You will need to add the following lines to your Cmake

SET(CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS} "-std=c++17 -pthread")
find_package(GSL REQUIRED)    
target_link_libraries(<folder name> GSL::gsl GSL::gslcblas)

Thank you for your time.

Answer 1

All threads always "have access" to the entire process space - i.e. all objects in the memory for that process.

In principle, nothing is required to "give X access" within the process.

Here:

post(pool, [&_A, &_dostuff] { _dostuff(); }); // post a task to the pool

you are overly verbose, which might indicate your confusion. _dostuff already captures the matrix, so you don't need that. Also, if you needed to capture the lambda you could do so by value (which is generally a safer default especially when threading):

post(pool, [=] { _dostuff(); });

Ironically, you could simply replace it with

post(pool, _dostuff);

Why Doesn't It Work?

Well it works. You just have a data race in your check - so anything can happen (https://en.wikipedia.org/wiki/Undefined_behavior).

To get a reliable check you might join the thread pool first:

#include <boost/asio.hpp>
#include <gsl/gsl_linalg.h>
#include <iostream>

namespace ba = boost::asio;

int main() {
    int SIZE = 3;                                  // small size
    gsl_matrix* _A = gsl_matrix_alloc(SIZE, SIZE); // 3x3
    gsl_matrix_set_all(_A, 0); // zero for reference

    auto _dostuff = [_A]() {
        gsl_matrix_set(_A, 0, 0, 20); // set (0, 0) to 20
    };

    ba::thread_pool pool(1); // one thread
    post(pool, _dostuff);    // post task
    pool.join();

    std::cout
        << gsl_matrix_get(_A, 0, 0)
        << std::endl;

    gsl_matrix_free(_A);
}

Prints

20

(also fixed the memory leak).

Side Notes:

You might want to use futures to synchronize on tasks. Also note that capturing the pointer by reference is unnecessary.